Case Study: Transforming BKMobil's Education Platform with AWS-Powered AI

In our project to enhance BKMobil's Metodbox education platform with an AWS-based AI solution, we experienced the excitement of developing an innovative assistant for both students and teachers. We are delighted to have been part of this collaboration and to contribute to BKMobil's vision for educational technology. The system we developed enables Turkish students and teachers to receive context-sensitive answers to their course-related questions, simplify homework management, and analyze student data.
 

Project Story

BKMobil decided to integrate AI technologies into their educational processes to further advance their already successful Metodbox platform. Our technical expertise combined with BKMobil's industry experience laid the foundation for a solution that makes a difference in educational platforms. The project's main objectives included increasing platform interactivity, providing more advanced search capabilities, and more effectively analyzing user data.
 

Technology Solution

The architecture we designed with BKMobil integrated seamlessly with their existing APIs while incorporating the AI capabilities offered by AWS services:
 

• Content Processing: Educational documents were processed using PDFPlumber running on Amazon EC2
• Vector Database: Processed content was stored as vectors in Amazon OpenSearch Service
• Application Layer: API services were provided using LangServe, FastAPI, and Uvicorn running on Amazon EC2
• Intelligent Workflows: Inter-agent workflows were created using LangGraph
• Performance Monitoring: AI model behaviors were tracked with ArizePhoenix
   

The AI architecture forming the backbone of our system utilizes both RAG (Retrieval-Augmented Generation) for contextual information and agents for performing complex tasks. This allows students and teachers to chat with the system in natural language to obtain information about course content, manage their homework, and access analytical data.

Infrastructure Stack

Amazon Bedrock: Served as our foundation model platform, enabling us to leverage Claude 3.7 Sonnet for generating contextual educational responses in Turkish and Amazon Nova for content guardrails, ensuring appropriate educational content delivery
Amazon Opensearch Service: Searching for relevant data in a RAG system is not a trivial task, we deployed Opensearch as our vector database. Amazon Opensearch Service helped us to deploy it and administer it without any difficulty. We used Opensearch as not only an approximate k-nn based vector search engine, but also for what it shines at: a distributed BM25 based search engine.
Amazon RDS: Provided reliable relational database services for chat history management and LLM communication logging, ensuring conversation continuity and providing a base for comprehensive analytics of student-AI interactions
Amazon EC2: Auto Scaling Groups managed our LangServe API layer and document processing workloads, automatically scaling during peak student usage periods, ensuring consistent response times
Amazon S3: Stored unstructured educational documents and processed embeddings with intelligent tiering, reducing storage costs while maintaining fast access for real-time queries.
 

Lessons Learned

During this project with BKMobil, our team gained several valuable experiences:

• API Integration: We experienced that automatic OpenAPI solutions alone are not sufficient for AI models' interaction with complex APIs, and that customized function wrappers are more effective.
• Content Processing: Instead of extracting data directly from PDF documents, we applied data extraction processes on labeled data provided from BKMobil's previous work. We had the opportunity to observe once again how data quality significantly improves RAG performance.
• Context Size Scalability: The increasing request context size with the growing number of APIs led to both cost and accuracy issues. We solved this challenge by dividing the system into smaller sub-agents.
   

This project resulting from our collaboration with BKMobil stands out as a successful application of modern AI technologies built on AWS infrastructure in the Turkish education system. Thanks to BKMobil's investment in educational technologies, students and teachers in Turkey can now benefit from an AI-powered assistant. We are happy to have been part of this transformation project and look forward to working with BKMobil on new projects.